Change patterns of precipitation anomalies and possible teleconnections with large-scale climate oscillations over the Yangtze River Delta, China

It is widely accepted that the climate is undergoing an intensive change worldwide, which has an important influence on the regional environment and humans (Wang et al. 2016). Among the climate variables, precipitation influences regional water resources, hydrological processes, ecology, agricultural production and living conditions directly. According to IPCC (2021), many regions do face not only a tendency toward drier conditions but also an increase in extreme precipitation intensities. In addition to these trends, meteorological conditions may show significant temporal variations over decadal or multi-decadal time scales. Therefore, a good understanding of these trends and variations is of importance to better support water resource management and engineering decision-making (Willems 2000; Huang et al. 2009; Martinez et al. 2012; Nalley et al. 2012; Duhan & Pandey 2013; Gocic & Trajkovic 2013; Ren et al. 2015; Thomas & Prasannakumar 2016; Wang et al. 2017).

As one of the most common meteorological elements, the characteristics of precipitation have been extensively studied. With the acceleration of the water cycle under the background of greenhouse warming, spatio-temporal patterns of precipitation tend to change in many parts of the world. Previous studies mainly used trend analysis methods, such as the Mann–Kendall test, Spearman's rho test and the Theil–Sen approach, and oscillation test methods, such as Morlet wavelet analysis, to detect the temporal variation of meteorological variabilities (Serrano et al. 1999; Tank & Können 2003; Sayemuzzaman & Jha 2014; Chen et al. 2015; Jones et al. 2015; Song et al. 2015; Yang et al. 2017). However, most of these analyses revealed the variation in the whole period, resulting in the lack of phase variation analysis. The application of these methods is also always restricted by some assumptions such as serial independence of time series (Yue et al. 2002; Wang et al. 2020). For example, the Mann–Kendall test, which can reflect the changing trend based on the ranks of measurements, was the most popular, because the sequences to be examined need not be subject to the characteristics of the same probability distribution (Kendall 1975). But its precondition (the data sequence was not autocorrelated) was always ignored (Tabari et al. 2014). Hydro-meteorological variables are always autocorrelated and magnify the significance of the trend. In addition, when discussing spatial patterns of precipitation change, administrative or watershed boundaries are often used. This means the spatial patterns presented by precipitation itself are ignored.

Hence, it is better to adopt a rather novel approach named the Quantile Perturbation Method (QPM) to avoid the above-mentioned assumptions in analyzing the change trend and mode of the precipitation. Analogous to the frequency–perturbation method, the QPM derives the research period into several baseline and corresponding activity periods with a sliding window. It needs fewer stringent assumptions for meaningful inferences from trend results of hydrological time series, compared with most of the commonly used non-parametric tests such as Mann–Kendall and Spearman (Yue & Wang 2004; Tabari et al. 2017; Xu et al. 2017). The method can not only analyze the return cycle of factors but also reveal the temporal variation of quantiles of hydro-meteorological variables in different periods (Ntegeka & Willems 2008; Willems 2013). At present, this method has been widely used in the research of long-term variation of hydro-meteorological factors in the world (Onyutha & Willems 2015, 2017; Xu et al. 2020). Meanwhile, as is a useful technique for analyzing series with complex periodical components, the Singular Spectrum Analysis (SSA) Method, a non-parametric technique of the time-series analysis method based on the principles of multivariate statistics, decomposes a given time series into a set of independent additive time series, so that each component in this set can be identified as either a trend, quasi-periodic component or noise (Zhang et al. 2017; Rubinetti et al. 2020). Until now, this method has been applied to analyze the temporal oscillations of hydro-meteorological variables (Marques et al. 2006; Hassani et al. 2010; Zhang et al. 2017). To reveal the regionalization of variables, the Principal Component Analysis (PCA) Method, which aims at extracting the coherent variations of dominant variables, is always applied (Huang et al. 2009; Wang et al. 2016).

Next to this, searching for the driver(s) of the variability in the precipitation anomalies is another important task as strong links have been identified between the teleconnection and lag relationship patterns and climate variability. It is well documented that precipitation variations were influenced by large-scale climate oscillations, such as El Niño-Southern Oscillation (ENSO), North Pacific index (NP) and Pacific Decadal Oscillation (PDO), and raised serious concerns (Mei et al. 2021). ENSO is a well-known climate oscillation to cause global climate variability by the cycle of warm and cold sea surface temperatures in the equatorial Pacific arising from complex interactions between the atmosphere and ocean, and has been commonly recognized to be closely related to the precipitation as well as the occurrences of drought and flood (Wang et al. 2021). The ENSO has a periodic behavior with bands of 2–3 and 4–7 years (Lall & Mann 1995). The link between precipitation anomalies and ENSO is statistically significant in parts of East Asia (Xu et al. 2004). The heavier precipitation in the northeastern Yangtze River Delta (YRD) of China is beneficial for obtaining persistent conditions by the plum rain belt during the El Niño warm periods (Ye et al. 2013). Extreme precipitation statistics were dominated by the ENSO in over 22% of the global land area (Cao et al. 2017; Sun et al. 2017). Researchers also concluded that the Arctic Oscillation (AO) has significant correlations with the precipitation during the winter and spring in large parts of central China (Gong & Wang 2003; Li & Leung 2013; He et al. 2017). The North Atlantic Oscillation (NAO) has had significant correlations between winter NAO and summer precipitation in north China in the last decades (Fu & Zeng 2005; Sung et al. 2006). Although the contemporaneous relation between precipitation and atmospheric oscillation patterns was revealed, the long memory of atmospheric anomalies also has a delayed influence on precipitation. The links offer considerable benefits for a good monthly and seasonal predictability of the relevant hazards for a few months ahead and need to be further estimated.

The YRD, located in East China, is the transitional zone between subtropical and temperate monsoon climate. The region is prone to drought and flood disasters, which hinder social progress and development. To address the aforementioned issues, the present study (a) emphasized the phase variation of precipitation anomalies and oscillation periods by the application of the QPM and SSA methods, (b) analyzed the spatial patterns of precipitation anomalies by the QPM and PCA methods, and (c) revealed the teleconnection and lag time between precipitation and the main large-scale climate oscillations, such as the AO, NAO, PDO, NP, Southern Oscillation Index (SOI), and ONI (Oceanic Niño 4 SST Index). This study will help to better understand the spatial and temporal patterns of precipitation anomalies in the YRD, as well as the teleconnections and the lag time between climate oscillations and precipitation. The results are anticipated to serve as a guideline for regional climate tendency and water resource assessment, and help production to adapt to climate change associated with large-scale climatic patterns.

Meanwhile, the YRD is one of the most important commodity grain production areas of China. The underlying surface covers a large portion of the cultivated land and requires a large amount of irrigation water. Apart from the status of an important agricultural base, the YRD is also one of the most developed areas in China. The three provinces and a municipality have a population of 227.14 million in 2021. The total 67.23% population urbanization rate supports a massive urban system of 152.69 billion populations. The massive urban system needs a lot of water for domestic and industrial use and inevitably causes water pollution. The industrial and agricultural production in the research region requires a large amount of water, but the variation of spatio-temporal distribution of precipitation poses a major threat.

Daily precipitation data were obtained from the National Meteorological Center of China (http://data.cma.gov.cn) from 1957 to 2016 in 43 stations (Figure 1 and Table 1). The quality of the data was checked and the missing data counted for less than 0.3%, and those data were interpolated through a simple linear regression model for daily precipitation with neighboring stations. To detect the possible influence of the multiple large-scale climate oscillations on the regional precipitation, monthly data of the climatic oscillations, such as AO, NAO, PDO, NP, SOI and ONI, were collected from the datasets (apdrc.soest.hawaii.edu/projects/monsoon and www.esrl.noaa.gov).

To detect the temporal variation patterns of precipitation, a rather novel approach called QPM was applied in this article (Ntegeka & Willems 2008; Willems 2013). The method compares the long-term baseline period value quantiles with those of a selected sub-period. The latter is taken for any particular block (subseries) of interest. This method was taken to detect anomalies of annual and seasonal total precipitation in 1957–2016 in the YRD.

Anomaly values, called perturbation factors, are calculated by following steps. In the first step, subseries of annual and seasonal total precipitation are derived from the full-time series. The length of the block period was selected according to the baseline period of 1957–2016. In general, a shorter length of the block period can reflect the disturbance intensity of precipitation, while a longer length can reflect the change trend. For this study, block periods of 5, 10 and 15 years with a sliding window of 1 year were tried to find the best one. For example, when the block period of 5 years with a sliding window of 1 year was selected, the following subseries are considered: 1957–1961, 1958–1962 and 1959–1963. In the second step, the time-series values are put in a descending order for each block period, such that they can be related to empirical return periods BL/i, where BL is the length of block series and i the rank number (i=1 for the highest value). After ranking, the values correspond with quantiles x(BL), x(BL/2),… , x(BL/i), …, where x(BL/i) is the quantile with the empirical return period BL/i. In the third step, perform steps 2 and 3 for the baseline period. In the next step, the quantiles of the block periods and baseline periods with a similar return period are compared. In the final step, the perturbation factor on the quantiles is calculated as a change ratio of the block period data and those of the baseline period.

To test the statistical significance of the anomaly values (perturbation factors), the values in the full-time series at each site were randomly resampled to make a new series with a different sequence, and the anomalies are recalculated for the resampled series based on the QPM. The anomaly calculations were repeated 1,000 times, leading to 1,000 anomaly values for each block period. After ranking the 1,000 anomaly factors, the 25th and 975th values were defined as the 95% confidence intervals for each block period. When the perturbation results were located outside the interval, this denotes the significant negative and positive anomalies.

Further analysis of the regional total precipitation by means of SSA (Table 2) shows that the 10 leading SSAPCs reflect the temporal variations of precipitation well at the annual scale and in the flood and non-flood seasons; the total variance explained by these SSAPCs reaches a value of approximately 70%.

Similar to the annual precipitation, the main oscillations of precipitation in the flood season are 2 years (for SSAPC-3, SSAPC-4 and SSAPC-5), 7–11 years (for SSAPC-1 and SSAPC-2), and 3–4 years (for SSAPC-6) (Figure 5(g), 5(i) and 5(k)), explaining 26.26, 35.59 and 6.04% of the total variance. Among these oscillations, the 2 and 7–11 years periods concentrate from the late 1970s to the beginning of the 1990s and from the 1950s to the 1980s, respectively. In the non-flood season, the oscillation period of 2 years is reflected by PSSACs-1, SSAPC-2 and SSAPC-5 and intensified in the 2010s (Figure 5(m), 5(o) and 5(p)). The 7–8 years one, which is reflected by SSAPC-3 and SSAPC-4, decreased gradually. The 3–4 year periods (for SSAPC-6, SSAPC-7 and SSAPC-8) do not show obvious temporal changes.

Though the annual and seasonal precipitation showed an insignificant trend in most of the stations, attention should be paid to the fluctuation processes of precipitation. Thus, the precipitation anomalies in all the meteorological stations were also detected by the QPM (Table 3). On the annual scale, significant negative anomalies were found in 23, 13 and 12 stations (53.49, 30.23 and 27.91% of the total) around the years 1968, 1978 and 2005, respectively. On the other hand, significant positive anomalies were found in 7, 6, 10 and 18 stations (16.28, 13.95, 23.26 and 41.86% of the total) around the years 1976, 1991, 1998–2002 and 2014, respectively. In the YRD, large-scale extremely dry events mainly happened in the late 1960s, the late 1970s and the 2000s assessed by the SPI method (Wu & Xu 2020). Meanwhile, several similar periods with significant positive/negative anomalies were also detected in flood and non-flood seasons. Thus, we can conclude that the similar significant negative and positive precipitation anomalies emerged in some stations at the same year.

To explore the change patterns of precipitation and their spatial distribution characteristics, the PCA method was adopted. The results of the PCA applied to the QPM results which showed that the first two PCAPCs explain more than 60% of the total variance (Table 4). Note that this variance refers to both the trend and the temporal anomalies because it was based on the QPM. Spatially, two dominant geographic sub-regions can be identified from that analysis. The temporal variation of anomalies for each of these sub-regions was reflected by the time series of corresponding PCAPC.

Figure 6(c) and 6(d) also clearly showed much higher values in the southern and northern YRD, where it is attributed to the first and second components of precipitation anomaly in the flood season, respectively (Figure 6(c) and 6(d)). Time series of variations for the first two PCAPCs demonstrated similar patterns for the flood season, which was due to the high relative contribution rate to the annual precipitation (Table 1). The similar temporal patterns and temporal variation of the PCAPCs further confirmed the contribution of precipitation in the flood season to annual precipitation in the monsoon climate zone. In the non-flood season, the first PCAPC showed high values mainly in the area north of the ‘Shengsi-Pinghu-Hangzhou-Tunxi’ line, whereas the opposite is true for the second PCAPC in general (Figure 6(e) and 6(f)). Time series for the first PCAPC illustrated high precipitation anomalies mainly for the mid-1960s and the 1970s, and for the late 1980s to the mid-2000s in the northern region (Figure 7(c)). Low anomalies were mainly found around the late 1960s, the 1980s and the 2010s.

Compared to AO and NAO, PDO had much stronger influence on the precipitation. The significant correlations were found in more than 25% of the total stations in March and July (Table 5). These stations were mainly located in the southeast region of YRD. The precipitation was also influenced by PDO in part of the mid-north region in September (Figure 8(c)). Meanwhile, NP also had a strong influence on precipitation in certain months. The strongest influence was detected in February, with the significant positive correlations happening in 33 of the total 43 (76.74%) meteorological stations (Table 5). In addition, 16 of the 43 (37.21%) stations also had strong positive correlations in October. These stations were mainly concentrated in the north region (Figure 8(d)). Interdecadal variation in the consecutive cloudy-rainy event frequency varies well in-phase with PDO in southern China through two physical processes. On the one hand, the anomalous PDO is associated with the anomalous North Pacific Oscillation, whose westward extension is accompanied by anomalous East Asian westerly jet and West Pacific subtropical high. On the other hand, the eastward development process associated with the PDO variation can excite a Rossby wave train over the Northern Hemispheric mid-latitudes (Gu et al. 2021).

The correlation showed that SOI and ONI have strong teleconnection with precipitation in March, September, October and November (Table 5). Spatially, these stations at which precipitation was significant correlated with SOI and ONI were both mainly located in the southeast region in January and August, the mid-north region in September and populated in most of the stations in March (Figure 8(e) and 8(f)). In addition, the former correlations were also distributed in the north and west regions in October and the mid-north region in November. The latter relations were also found in southeast, northwest and west regions in October and dominated in most regions in November. Generally, the ENSO changes and flood/drought variation were significantly correlated at about 5-year and 10–12-year periods, which coincided with the oscillations of precipitation in the YRD detected by SSA, in the middle and lower reaches of the Yangtze River Basin (Jiang et al. 2006). ENSO episodes are in good teleconnection with floods/droughts in the Yangtze River catchment. Eastern Asian monsoons are influenced by ENSO through the strength of the subtropical high in the western Pacific region (Jiang et al. 2006; Zeng & Sun 2022).

Because of the natural evolution and human activities, the earth's surface has experienced intensive global warming since the 1970s, which interacts with the global water cycle. Both the increasing and decreasing trends of precipitation were detected by researchers in different regions of the world. Although precipitation can be regarded as a random event, long-term precipitation observations in an area show certain regular characteristics. In most parts of China, precipitation is mainly brought by the East Asia and Indian monsoons. Generally, the annual average total precipitation decreased from the southeast to northwest of China. Temporally, precipitation is concentrated in the flood season. What's more, the massive water resources are needed because of the developed economy and large population. Too much or too little precipitation has a huge impact on crop yields in agriculture, and flood control and water supplement in urban areas. Thus, it is necessary to detect the change modes and influences in depth for better water resource management and engineering decision-making.

This article provided an enhanced insight into the variation patterns of precipitation at 43 meteorological stations in the YRD from 1957 to 2016 and their relationship with the main large-scale climate oscillations. The main conclusions of the study were as follows:

• (1)

  Through the SSA, the main oscillations of precipitation were all found to be 2, 7–11 and 3–4 years in the annual and seasonal scales. Temporal and spatial patterns of precipitation anomalies were both detected in the annual scale, the flood season and the non-flood season. Spatially, significant anomalies were found at the annual scale in the mid-1960s and the beginning of the 1980s, and the flood season in the mid-1960s and the late 1970s. Anomalies were insignificant in the non-flood season. Spatially, two main patterns of precipitation anomalies were revealed based on PCA in the annual scale, the flood season and the non-flood season, respectively. Precipitation anomalies showed an increasing trend in the south region, but kept stable in the north region in the annual scale and the flood season. In the non-flood season, precipitation anomalies increased in both regions, and the southeast regions increased more intensively.

• (2)

  Among the selected large-scale climate oscillations, AO and NAO showed weak correlations with the precipitation, and the lag times were 0 and 3–5 months. PDO and NP have significant correlations with precipitation in March, July and September in the southeast regions and in February and October in the north region. The lag times were 0–3 and 0–2 months, respectively. The oscillations of SOI and ONI influenced the precipitation in spring and autumn in several regions. The lag times were 2–3 and 1–5 months, respectively.

The authors wish to thank the National Meteorological Information Center of the China Meteorological Administration and the U.S. Earth System Research Laboratory for offering the meteorological data. The authors also gratefully acknowledge the editors and anonymous reviewers for their constructive comments on the manuscript.

This research work was financially supported by the National Natural Science Foundation of China under Grant 42001025 and 42001014, the Belt and Road Special Foundation of the State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering under Grant 2021491211, the Research Program of Ningbo University under Grant 026-422002842, the Fund of Sustainable Urban Drainage Laboratory of Ningbo University under Grant 09, the Fund of Humanity and Social Science Youth Foundation of Ministry of Education of China under Grant 20YJCZH180, the Fund of Zhejiang Public Welfare Technology Research Project under Grant LGF21D010003 and the Ningbo Fan-3315 Plan under Grant 202002N3200.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.